3 ms·
This is true in general, but deep learning with lots of hyperparameters gives an advantage to organizations with clusters that can run a huge hyperparameter sea
by blt 8y ago
This is true in general, but deep learning with lots of hyperparameters gives an advantage to organizations with clusters that can run a huge hyperparameter search quickly.
- Zephyr314 8y agoSigOpt (YC W15) has an academic program that allows full access to their optimization platform for academic use [1]. Hundreds of academics around the world have used it and there are already dozens of papers that have been published that have benefited from it [2]. These techniques can be orders of magnitude more efficient than a standard hyperparameter search and really cut down the barrier to entry for these types of results. [1]: https://sigopt.com/solution/for-academia/ https://sigopt.com/solution/for-academia/ [2]: https://sigopt.com/research/publications/ https://sigopt.com/research/publications/